3.9 — Case study: a lead qualification agent 🖥️

Let’s build (on paper) a no-code agent that qualifies incoming prospects. The classic scenario: new lead → Airtable → AI qualifies → email.

🟢 In plain words — “qualifying a lead” simply means sorting a prospect: are they ready to buy (hot), interested but not in a hurry (warm), or unlikely (cold)? The agent does this sorting automatically and triggers the right follow-up — an email right away, or a reminder later.

The workflow, step by step (in Make or n8n):

1. Trigger — a new lead fills out a form on your site (via a webhook: an address that notifies your workflow as soon as an event happens).
2. Memory — Make/n8n creates a row in Airtable with their info (name, company, message).
3. Brain — the AI node (ChatGPT or Claude) reads the row and qualifies the lead (hot / warm / cold) thanks to a structured prompt (your 4 ingredients from Level 2!).
4. Observation — the score is written back to Airtable (the memory updates).
5. Action — if hot → the agent drafts and sends a personalized email; if cold → it schedules a follow-up in 7 days.

🔁 Recognize the ReAct loop (3.4)? Reason (qualify the lead) → Act (write to Airtable / send the email) → Observe (the result) → Loop. It’s a ReAct agent… without a line of code. Make/n8n = the loop; Airtable = the memory; the assistant = the brain; your Level 2 prompt = the instructions.

Caution

⚠️ Level 4 preview — already an alert: this agent sends emails to real customers → that’s a sensitive action. In Module 4, we’ll add a human validation point (HITL) and guardrails before any send. Power = guardrails.

✅ In short

• A no-code agent follows the pattern trigger → memory → AI brain → action.
• It’s a concrete ReAct loop: Make/n8n = the loop, Airtable = the memory, the assistant = the brain, your Level 2 prompt = the instructions.
• As soon as an action touches real customers, it becomes sensitive → guardrails and human validation (Module 4).